Anthropic signed a $35 billion cloud computing agreement with Lambda, the Nvidia-backed GPU infrastructure provider, tied to a 350-megawatt artificial intelligence datacenter in Texas. The deal is structured as a long-term capacity commitment, not a capital purchase, meaning Anthropic secures compute without building or financing the facility itself. Lambda carries the construction and operational risk. Nvidia's involvement extends beyond chip supply—the company is providing project-level financing and engineering coordination for the buildout, effectively underwriting the infrastructure stack required to train foundation models at Anthropic's projected scale through 2030.

The deal is part of a new financing architecture Anthropic has been assembling since late 2024. Instead of raising equity to buy GPUs and lease colocated space, the company is negotiating offtake agreements with cloud providers who finance the datacenter construction in exchange for long-term compute commitments priced at a discount to retail cloud rates. The Lambda contract follows a similar structure to Anthropic's earlier agreements with CoreWeave and Oracle, though this is the largest single commitment disclosed to date. The 350 MW facility—roughly equivalent to the power draw of a mid-sized semiconductor fab—will be built on land controlled by Hut 8, a publicly traded Bitcoin mining operator pivoting into AI infrastructure hosting. Hut 8's Texas site was selected for its existing high-voltage electrical infrastructure and direct access to ERCOT grid capacity, both critical for datacenter projects requiring sustained multi-hundred-megawatt power delivery.

This structure changes the capital efficiency calculus for frontier AI labs. Anthropic avoids the $10-15 billion in upfront capital expenditure required to self-finance a cluster of this scale, and it shifts utilization risk to Lambda, which must fill unused capacity or absorb the cost of idle infrastructure. For Lambda, the arrangement is a bet on Anthropic's revenue trajectory and model training roadmap—if Anthropic's enterprise API revenue grows as projected, the contract locks in a marquee anchor tenant and justifies the construction debt. For Nvidia, the financing role deepens its strategic position: the company is no longer just a chip vendor but a co-investor in the success of its largest customers. The deal also signals that Nvidia is willing to provide vendor financing to accelerate AI infrastructure buildouts, a move that could become standard for hyperscale AI projects where traditional project finance is unavailable or prohibitively expensive.

Operators should watch for two follow-on developments. First, whether Lambda secures debt financing from infrastructure-focused lenders or converts the Nvidia backing into a formal credit facility—project announcements of this scale typically require $5-7 billion in construction and equipment debt, and the lending terms will indicate how much risk the credit markets are willing to take on AI-specific datacenter projects. Second, whether Anthropic's model holds under stress—if training runs take longer than expected or if API revenue growth slows, the company may struggle to meet its capacity commitments, creating downstream pressure on Lambda's debt service and Nvidia's exposure. The Texas facility is expected to begin partial operations in Q2 2026, with full capacity online by mid-2027.

The Hut 8 site selection is the cleaner story. Bitcoin miners built for 200+ MW continuous load and direct ERCOT interconnects, which is exactly what AI datacenters need and exactly what new construction projects struggle to permit.